def train_cnn_model(width, height, num_samples, needs_split=False):
"""
This trains a cnn model and outputs the model to the specified location below
:param width: width of image
:param height: height of image
:param num_samples: total number of samples that are being split
:param needs_split: if the samples are split into the output folder or not
"""
start = time.time()
if needs_split:
splitfolders.ratio("assets/cnn_images_5m/", output="assets/cnn_images_5m/output",
seed=0, ratio=(0.8, 0.1, 0.1), group_prefix=None)
model_metrics = ['accuracy', metrics.BinaryCrossentropy(), metrics.Precision(),
metrics.Recall(), metrics.BinaryAccuracy()]
img_width, img_height = width, height
train_data_dir = 'assets/cnn_images_5m/output/train'
val_data_dir = 'assets/cnn_images_5m/output/val'
test_data_dir = 'assets/cnn_images_5m/output/test'
epochs = 10
batch_size = 32
validation_steps = 6494 // batch_size # number of validation samples available
classes_num = 2
nb_filters1 = 16
nb_filters2 = 32
conv1_size = 4
conv2_size = 2
pool_size = 2
model = Sequential()
model.add(Convolution2D(nb_filters1, conv1_size, conv1_size, input_shape=(img_width, img_height, 3)))
model.add(Activation("relu"))
model.add(MaxPooling2D(pool_size=(pool_size, pool_size)))
model.add(Convolution2D(nb_filters2, conv2_size, conv2_size))
model.add(Activation("relu"))
model.add(MaxPooling2D(pool_size=(pool_size, pool_size)))
model.add(Flatten())
model.add(Dense(256))
model.add(Activation("relu"))
model.add(Dropout(0.5))
model.add(Dense(classes_num, activation='softmax'))
model.summary()
model.compile(loss='binary_crossentropy',
optimizer='adam',
metrics=model_metrics)
train_datagen = ImageDataGenerator()
test_datagen = ImageDataGenerator()
# buy is 0, hold is 1
train_generator = train_datagen.flow_from_directory(
train_data_dir,
target_size=(img_height, img_width),
batch_size=batch_size,
class_mode='categorical')
validation_generator = test_datagen.flow_from_directory(
val_data_dir,
target_size=(img_height, img_width),
batch_size=batch_size,
class_mode='categorical')
log_dir = './tf-log/'
tb_cb = callbacks.TensorBoard(log_dir=log_dir, histogram_freq=0)
cbks = [tb_cb]
history = model.fit(
train_generator,
batch_size=batch_size,
steps_per_epoch=num_samples // batch_size,
epochs=epochs,
validation_data=validation_generator,
callbacks=cbks,
validation_steps=validation_steps)
target_dir = './models/'
if not os.path.exists(target_dir):
os.mkdir(target_dir)
model.save(f'assets/models/joey_cnn_intraday/cnn_model_5m_{epochs}epochs_{classes_num}classes.h5')
model.save_weights(f'assets/models/joey_cnn_intraday/cnn_weights_5m_{epochs}epochs_{classes_num}classes.h5')
with open(f'assets/models/joey_cnn_intraday/history_5m_{epochs}epochs_{classes_num}classes.pkl', 'wb') as f:
pickle.dump(history.history, f)
# Calculate execution time
end = time.time()
dur = end - start
if dur < 60:
print("Execution Time:", dur, "seconds")
elif 60 < dur < 3600:
dur = dur / 60
print("Execution Time:", dur, "minutes")
else:
dur = dur / (60 * 60)
print("Execution Time:", dur, "hours")
Above is a CNN model schema example seen here. The first part consists of Convolutional and max-pooling layers, while the second part consists of the fully connected layer, which performs non-linear transformations of the extracted features and, in this case, acts as the classifier. Two types of outputs here can be either a softmax or a sigmoid layer, which will tell whether the image is of a cat or not, or probabilty of the image being a cat.
